UMI-3D Dataset and Models

---

Overview

This repository provides datasets, pretrained models, and raw recordings for UMI-3D.

Each task includes:

• A Zarr dataset for policy training
• Pretrained policy weights
• Raw rosbag recordings for reproducibility

---

Task 1 — Cup Arrangement

• 📦 Zarr Dataset
  https://huggingface.co/datasets/ZIMBOT/umi-3d-dataset-cup

• 🤖 Pretrained Policy
  https://huggingface.co/ZIMBOT/umi-3d-policy-cup

• 🧾 Raw Rosbag Data
  https://pan.baidu.com/s/1ukL6oDFspRRjbqr6Vdy5VQ?pwd=kyg6

---

Task 2 — Curtain Pulling

• 📦 Zarr Dataset
  https://huggingface.co/datasets/ZIMBOT/umi-3d-dataset-curtain

• 🤖 Pretrained Policy
  https://huggingface.co/ZIMBOT/umi-3d-policy-curtain

• 🧾 Raw Rosbag Data
  https://pan.baidu.com/s/1-ZTBTXOFqplpWlRUnBSAIQ?pwd=4m3j

---

Task 3 — Door Opening and Cup Arrangement

• 📦 Zarr Dataset
  https://huggingface.co/datasets/ZIMBOT/umi-3d-dataset-door-cup

• 🤖 Pretrained Policy
  https://huggingface.co/ZIMBOT/umi-3d-policy-door-cup

• 🧾 Raw Rosbag Data
  https://pan.baidu.com/s/1QVurpNRc9u0ERTpR_iixhg?pwd=k5nb

---

Notes

• The datasets are generated using the full UMI-3D pipeline (Hardware → SLAM → Data Processing).
• Raw rosbag files are provided for full reproducibility and reprocessing.
• All datasets are compatible with the UMI-3D Policy framework.

---

Citation

If you find this work useful for your research, please consider citing:

@misc{wang2026umi3dextendinguniversalmanipulation,
  title={UMI-3D: Extending Universal Manipulation Interface from Vision-Limited to 3D Spatial Perception},
  author={Ziming Wang},
  year={2026},
  eprint={2604.14089},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2604.14089}
}